QuestBench - Mean Score: leaderboard

QuestBench (2026) holds 256 expert-level deep research questions in Chinese from 14 humanities and social-science domains, designed and adversarially reviewed by students and answered with web search and visit tools. This board is the mean graded score from 0 to 100; higher is better.

Metric: Mean score (0-100) against question-specific grading criteria over the 256 Chinese deep research questions, with identical web search and visit tools and a nominal 50-call budget per question; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.567.12#54
2Claude Opus 4.757.79#85
3Gemini 3.1 Pro (Preview)43.07#99
4GLM-5.142.3#198
5DeepSeek V4 Pro38.24#196
6Kimi K2.637.74#192
7Qwen 3.6 Plus36.05#238
8MiMo-V2.5-Pro32.05#284
9Kimi K2.530.48#266
10MiniMax-M2.729.29#539
11Seed-2 Pro22.89
12Seed-1.8 Pro22.07
13DeepSeek V3.214.58#419

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=questbench-mean-score · How It Works · Data refreshed daily, snapshot 2026-11-21.